reminds me of this one:

If you torture data sufficiently, it will confess to almost anything. -Fred
Menger, chemistry professor (1937- )

PS: I think that copy-paste from me was a bit of quote mining on my part, but never mind :)

But knowing how a drug works makes for an infinitely more effective trial, especially when checking for negative side-effects …

For example, knowing that paracetamol is metabolized by the liver, and that the real drug that works on the brain is a metabolite of it, calls for special investigations about possible liver damage.

And since knowing how a drug works improves the effectiveness of the controlled trial so much – investigating about how it, or at least trying, is mandatory as well, in my opinion.

Anyway, out of metaphors – Oracle is a “bit” less complex than a human (or even an algae). You don’t need a PhD and a full research team when conducting controlled trials … just average analytic skills, sqlplus and a lot of sweating.

Not really the right way of thinking about it. The purpose of the trials is to eliminate errors and ensure that the step from “simple correlation” to “probable cause” is justified. (And to check for threatening side-effects of course –

Your use of the phrase “(no causation)” is also not entirely appropriate – it is reasonable to recognise causation without understanding (or having a complete understanding of) mechanism. In many cases understanding of mechanism comes later – and results in refinement of treatment or predications of side effects that need to be addressed.

It makes for a good read on this notion in regard to the Oracle DB and bulk binds.
http://www.oracle.com.cn/archiver/?tid-56543.html

For those regularly reading information by people such as Ben Goldacre, Stephen J. Gould; and hence potentially Richard Dawkins, PZ Myers & Phil Plait – I’m sure you’re all familiar with the dangers of data mining (and quote mining) and have seen plenty of non-Oracle related examples.

For anyone who reads a newspaper – not necessarily a tabloid, I think can see that statistics can be misrepresented to (attempt to) say whatever the author would like to portray.

My 5 cents.

>> Surely many of them will be different after I change a parameter or a query.

And yes, they do!

It’a an iterative process, start with low hanging fruit, apply, rinse and repeat . . .

According to my doctor (I took this stuff to quit smoking), that’s only the theory.

There is no proof, and they still have no idea about how it works (no causation).

>> Drug company verified findings in a controlled trial

And that is to confirm the correlation. As you correctly noted, that’s all that medical science requires.

It is extraordinary how even the more respectable papers and news programs will produce headlines and soundbites that are clearly idiotic cherry-picking, compression, and hyping of cautiously stated results from careful investigations.

Notice that the drug was first approved by the FDA in December 1985, withdrawn in 1986, and re-introduced in 1989; but it wasn’t until 1997 – more than 11 years after the first release, and 8 years after the re-release – that it was approved as an aid to quit smoking and sold as Zyban.

There’s probably a few years of testing in between – and it’s quite likely that most of that testing would be looking for harmful side-effects.

Then – in 2006 – the drug was also approved for as a treatment for SAD (seasonal affective disorder).

It would be interesting to know how much of the time lag went into:

a) initial observation of the correlation
b) testing for causality
c) testing for side effects
d) research into why it works.

(Obviously this isn’t intended to suggest non-overlapping time-intervals).

Thats pretty neat.

The research probably went like this:
1) Findings were reported from doctors and patients about this great side-effect of Wellbutrin.
2) Drug company verified findings in a controlled trial.
3) Researches found the reason it works.

Note that step 3 is optional, but step 2 is mandatory for turning an interesting idea into a medicine.
Controlled experiments is what turns a correlation into a cause by eliminating other possible explanations such as chance, the fact that non-depressed people find it easier to quit smoking, etc.